Episode 2.1: Install Claude Code
Install Claude Code, Anthropic's command-line agent, on Windows (or a Mac), sign in, start it in your project folder, and check each step before you let it speed up.
Downloads
AI for Research Efficiency, episode 2.1, with Douglas Hutchings. Narration: an AI-generated voice (ElevenLabs).
Commands as of 2 October 2026, from Anthropic's documentation (https://code.claude.com/docs/en/setup (opens in a new tab)). They change: check the documentation before you run them.
1. Install
Windows
Windows PowerShell:
irm https://claude.ai/In Command Prompt instead of PowerShell:
curl -fsSL https://claude.ai/Mac
Mac (Terminal):
curl -fsSL https://claude.ai/Other ways: WinGet: winget install Anthropic.ClaudeCode (does not update itself). Homebrew (Mac): brew install --cask claude-code. npm (needs Node.js 22 or later): npm install -g @anthropic-ai/claude-code.
2. Open a new terminal, and check
claude --version3. Start it in your project folder
cd agent-workclaudeThe first time, Claude Code asks whether you trust the folder. Since version 2.1.289 (3 October 2026) the prompt lists No, exit first: move to Yes, I trust this folder with the arrow keys, then press Enter.
4. Your account
A Claude plan (Pro, Max, Team or Enterprise), or an Anthropic Console account (API usage billing). The free claude.ai plan does not include Claude Code. An API key in ANTHROPIC_API_KEY is used instead of your plan once you approve it, and billed at API rates.
5. Check each step, then speed up
Claude Code now starts in auto mode: a second model reviews its actions instead of you. While you learn, press Shift+Tab once for manual mode (it asks before it edits a file, runs a command, or reaches the network), or start it with:
claude --permission-mode manualWhen you trust it with a task, press Shift+Tab until the status line says auto mode. Anthropic: "Auto mode reduces permission prompts but does not guarantee safety."
If something goes wrong
| You see | What to do |
|---|---|
| 'claude' is not recognized | The install folder isn't on your PATH yet: open a new PowerShell window and try again. |
| 'irm' is not recognized | You are in Command Prompt, not PowerShell: use the Command Prompt line above, or open PowerShell. |
| Claude Code does not support 32-bit Windows | You opened Windows PowerShell (x86): open the ordinary Windows PowerShell. |
Your data
On Free, Pro and Max accounts, Anthropic trains future models on your Claude Code sessions only if the setting that allows it is on; under commercial terms (Team, Enterprise, the API, Claude for Education) it does not, by default. Check your campus's rules for research data before you point an agent at it.
Sources
- Anthropic: Claude Code docs: Choose a permission mode (opens in a new tab), read 2 October 2026
- npm, Inc.: npm registry metadata for the three packages (opens in a new tab), read 2 October 2026
- Anthropic: Claude Code docs: Advanced setup (system requirements, install, update, uninstall) (opens in a new tab), read 2 October 2026
Corrections
None so far. If you find something wrong, email doug,@douglashutchings.com.
Transcript
Every spoken line, by chapter
Claude Code
In a few minutes, Claude Code will be working in your project folder, and you'll know how to keep it in check.
AI for Research Efficiency. Episode 2.1: Install Claude Code.
Last time: a terminal, a project folder, Git and Python, and an account.
In this episode: install Claude Code, sign in, run it in your project folder, and check each step before you speed up.
Install
Step one: in PowerShell, paste this line: irm https://claude.ai/install.ps1 | iex
On a Mac, use the second line, in Terminal. Both lines are on the companion page.
You don't need administrator rights.
Check the version
Step two: close the terminal, open a new one, and type claude --version. You should see a version number.
If PowerShell says claude is not recognized, open another PowerShell window, and try again.
Start in your folder
Step three: go into your project folder with cd agent-work, and type claude.
Always start it in a project folder, never in your home folder.
Sign in
Step four: the first time, it asks you to pick a text style, then how to sign in: a Claude plan, or an Anthropic Console account for an API key.
Your browser opens. Sign in there, and the terminal says: login successful.
Trust the folder
Step five: it reminds you that Claude can make mistakes, and that you're responsible for reviewing what it does.
Then it asks whether you trust this folder. It's your own project folder, so: yes.
Check each step
Step six: Claude Code now starts in auto mode, where a second model approves most of its actions. While you learn, press Shift and Tab once, for manual mode.
In manual mode, it stops and asks you before it edits a file, runs a command, or reaches the network.
A first question
Step seven: ask it something. What's in this folder?
It reads the notes and the data, and answers in plain words. To open the Word drafts, it needs to run a converter: that's a command, so it asks you first.
Speed up
Step eight: when you trust it with a task, press Shift and Tab until the status line says auto mode. It works without stopping, and much faster.
Anthropic's own advice: auto mode reduces permission prompts, but it doesn't guarantee safety.
Try it
Pause now: install Claude Code, open a new terminal, and check the version.
Review and next
So: one line to install, a new terminal, your project folder, sign in, trust the folder, manual mode first, then auto mode when you trust it.
Next: a worked example. Agents read a whole literature folder in parallel, and check every claim.